A train positioning system and method

Abstract

This invention discloses a train positioning system and method, relating to the field of train positioning technology. The method includes: continuously acquiring train operation data; detecting and identifying the virtual transponder number on a trackside sign; aligning the image and point cloud data along the railway line from spatial and temporal dimensions to obtain corrected point cloud information of the target area; calculating the distance from the train to the trackside sign and a first confidence interval based on the corrected point cloud information; converting this into the train's on-track position and a second confidence interval; predicting the train's passing time; and triggering the virtual transponder when the train passes the trackside sign. In the existing train control system architecture, by detecting the trackside sign and extracting the virtual transponder number, combined with point cloud data and electronic map information, the absolute train positioning function is achieved by triggering the virtual transponder when the train passes the trackside sign.

Description

Technical Field

[0001] This invention relates to the field of train positioning technology, and more specifically to a train positioning system and method. Background Technology

[0002] Currently, continuous, accurate, and reliable positioning is a prerequisite for safe train control and a crucial function of train operation control systems. Current train positioning systems generally employ a combination of onboard wheel and axle speed sensors and transponder positioning technology to achieve seamless and reliable positioning. The wheel and axle speed sensors continuously provide the train's relative displacement; the transponder positioning technology specifically involves the onboard equipment receiving the transponder signal via a transponder receiving unit as the train passes a transponder installed on the track. The demodulated information is then transmitted to the onboard main control unit, which obtains the train's absolute position by parsing the transmitted message. Therefore, the position provided by transponder positioning technology is a discrete absolute position, and train positioning systems generally use a one-dimensional coordinate system based on the distance to the previous transponder to describe this position.

[0003] However, this approach makes the positioning function highly dependent on the density of transponders along the line and the reliability of the transponder equipment. The complex and harsh natural environment along railway lines, such as rain, dust storms, extreme temperatures, temperature differences, and humidity, will all reduce the reliability of physical transponders, thereby increasing equipment maintenance costs and construction difficulty. Furthermore, long-distance, low-population-density railway sections, such as deserts and high-altitude mountainous areas, present challenges in transponder maintenance.

[0004] Therefore, how to provide a train positioning system and method that, without changing the existing system architecture of the train control system, partially replaces the physical transponder for positioning in complex line sections and achieves the absolute positioning function of the train is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] In view of this, the present invention provides a train positioning system and method that addresses the difficulties in maintaining physical transponders in harsh natural environments and low-population-density line sections. Without changing the existing system of the train control system, the present invention partially replaces physical transponders for positioning in complex line sections with virtual transponders, thereby achieving absolute train positioning.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a train positioning method, comprising:

[0007] Continuously collect train operation data, which includes railway line image data, railway line point cloud data, and train operation pulse data;

[0008] Based on the image data along the railway line, the virtual transponder numbers on the trackside signs are detected and identified.

[0009] From the spatial and temporal dimensions, the image data and point cloud data along the railway line are aligned to obtain the point cloud information of the corrected target area. Based on the point cloud information of the corrected target area, the distance from the train to the trackside sign and the first confidence interval are calculated.

[0010] Based on electronic map data, the distance from the train to the trackside sign and the first confidence interval are converted into the train's on-track position and the second confidence interval. The train's passing time is predicted, and a virtual transponder is triggered when the train passes the trackside sign; otherwise, the trackside sign detection is repeated.

[0011] Preferably, detecting and identifying the virtual transponder number on the trackside sign includes:

[0012] A supervised learning-based target detection method is used to detect target areas containing trackside signs.

[0013] The target area is identified using OCR character recognition technology to obtain the virtual transponder number.

[0014] Preferably, image point cloud alignment is performed on the railway line image data and the railway line point cloud data from both spatial and temporal dimensions, including:

[0015] From a spatial perspective, the pixel coordinates of the target area are projected onto the unit sphere of the binocular camera coordinate system based on the intrinsic parameters of the binocular camera; the target area is then transformed from the binocular camera coordinate system to the lidar coordinate system based on the extrinsic parameters of the binocular camera to obtain the target point cloud region.

[0016] From a temporal perspective, the image data and point cloud data along the railway line are time-aligned. Based on the sampling time difference between the image frame and the point cloud frame, the target point cloud region is corrected to obtain the point cloud information of the corrected target region.

[0017] Specifically, image data along the railway line is acquired through the binocular camera.

[0018] Preferably, calculating the distance from the train to the trackside sign and the first confidence interval includes:

[0019] The point cloud information of the target area is filtered by reflection intensity, and the filtered point cloud information is fitted by plane in combination with the actual size of the trackside sign.

[0020] Calculate the centroid coordinates of the planar point cloud, where the centroid coordinates of the planar point cloud represent the position of the trackside sign, and obtain the distance between the train and the trackside sign;

[0021] The first confidence interval is determined based on the inherent ranging error, plane fitting error, centroid calculation error of the lidar equipment, and the installation error of the trackside sign in the direction perpendicular to the track line.

[0022] Specifically, point cloud data along the railway line is acquired through the lidar device.

[0023] Preferably, the on-track position of the virtual transponder is obtained from the electronic map based on the virtual transponder number, and the on-track position of the train is obtained by combining the distance between the train and the virtual transponder.

[0024] Preferably, the data is based on the train operation pulse data and the distance between the train and the trackside sign.

[0025] A state equation is constructed using the on-track distance between the train and the virtual transponder as the state variable to predict the on-track distance between the train and the virtual transponder; at the same time, the first confidence interval is converted into a second confidence interval.

[0026] And trigger the virtual transponder when the state vector is less than zero and the trackside sign detection fails;

[0027] Wherein, the first confidence interval is the confidence interval for measuring the distance between the train and the trackside sign;

[0028] The second confidence interval is the confidence interval of the on-orbit distance between the train and the virtual transponder.

[0029] Preferably, the state equation is expressed as:

[0030]

[0031] Where, x k Let u be the on-orbit distance between the train and the virtual transponder at time k. k y represents the distance the train travels. k Let α be the distance between the train and the trackside sign, and β be the angles between the horizontal and vertical directions of the laser radar device, respectively. k n k These are the input noise from the wheel axle speed sensor and the measurement noise from the train's on-track position, respectively.

[0032] Preferably, the second confidence interval consists of two parts: a random part and a fixed part;

[0033] The random part is the confidence level obtained by solving the state equation to get the state vector;

[0034] The fixed components include the track straightness assumption error and the installation deviation of the trackside sign along the track direction;

[0035] The error portion of the trajectory straight line assumption is as follows:

[0036]

[0037] In the formula, r is the minimum radius of curvature of the track, α and β are the angles between the horizontal and vertical directions of the laser radar device in polar coordinates, and L represents the distance between the train and the trackside sign.

[0038] Finally, the second confidence interval is expressed as:

[0039]

[0040] In the formula, dev is the installation deviation of the trackside sign along the track direction, and P is the confidence level of the state vector obtained by solving the state equation.

[0041] Preferably, the optimal estimate of the state equation is obtained by Kalman filtering, wherein the prediction step is calculated as follows:

[0042]

[0043] Where Q represents the noise of the wheel axle speed sensor;

[0044] The Kalman gain is calculated as follows:

[0045]

[0046] Where R is the noise level measured between the train and the trackside sign.

[0047] The updated state vector and confidence are obtained as follows:

[0048]

[0049] In the formula, Let be the prior covariance matrix of the state vector at time k; Let be the posterior covariance matrix of the state vector at time k-1; This is the prior estimate of the state vector at time k; This is the posterior estimate of the state vector at time k-1; K is the posterior estimate of the state vector at time k; k The Kalman gain at time k; Let be the posterior covariance matrix of the state vector at time k.

[0050] Preferably, a train positioning system includes:

[0051] The sensor module is used to collect image data, point cloud data, and train operation pulse data along the railway line.

[0052] Electronic maps are used to store track geographic information and virtual transponder data required for train positioning;

[0053] The integrated positioning unit is used to receive and process the railway line image data, railway line point cloud data and train operation pulse data, obtain the distance between the train and the trackside sign, and trigger the virtual transponder when passing the trackside sign.

[0054] As can be seen from the above technical solution, compared with the prior art, the present invention discloses a train positioning system and method, including: continuously collecting train operation data, detecting and identifying the virtual transponder number on the trackside sign; aligning the image point cloud data and point cloud data along the railway line from spatial and temporal dimensions to obtain the corrected point cloud information of the target area, and calculating the distance from the train to the trackside sign and a first confidence interval based on the corrected point cloud information of the target area; converting this into the train's on-track position and a second confidence interval, predicting the train's passing time, and triggering the virtual transponder when the train passes the trackside sign. Under the existing train control system architecture, by detecting the trackside sign and extracting the virtual transponder number, combined with point cloud data and electronic map information, the absolute positioning function of the train is achieved by triggering the virtual transponder when the train passes the trackside sign. Attached Figure Description

[0055] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0056] Figure 1 This is a schematic diagram of the train positioning system provided in an embodiment of the present invention.

[0057] Figure 2 This is a schematic diagram of the installation of a binocular camera and a lidar device provided in an embodiment of the present invention.

[0058] Figure 3 This is a schematic flowchart of a train positioning method provided in an embodiment of the present invention.

[0059] Figure 4 This is a schematic diagram of the target detection and recognition method provided in an embodiment of the present invention.

[0060] Figure 5 This is a schematic diagram illustrating the change in the horizontal angle between two sampling times provided in an embodiment of the present invention.

[0061] Figure 6 This is a schematic diagram illustrating the change in the vertical angle between two sampling times provided in an embodiment of the present invention.

[0062] Figure 7 A flowchart for calculating the distance between a train and a trackside sign, provided for an embodiment of the present invention.

[0063] Figure 8 This is a schematic diagram of trajectory straight line assumption error compensation provided in an embodiment of the present invention. Detailed Implementation

[0064] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0065] This invention discloses a train positioning method, comprising: continuously collecting train operation data, including railway line image data, railway line point cloud data, and train operation pulse data; detecting and identifying virtual transponder numbers on trackside signs based on the railway line image data; aligning the railway line image data and railway line point cloud data in spatial and temporal dimensions to obtain corrected point cloud information of a target area, and calculating the distance from the train to the trackside sign and a first confidence interval based on the corrected point cloud information of the target area; converting the distance from the train to the trackside sign and the first confidence interval into the train's on-track position and a second confidence interval based on electronic map data, predicting the train's passing time, and triggering the virtual transponder when the train passes the trackside sign; otherwise, re-detecting the trackside sign. This invention achieves absolute train positioning by detecting trackside signs and extracting virtual transponder numbers, combining point cloud data and electronic map information, and triggering the virtual transponder when the train passes the trackside sign.

[0066] In one specific embodiment of the present invention, a train positioning method, such as... Figure 3 As shown, the main steps include:

[0067] S1. After installing and fixing the binocular camera, lidar device and wheel speed sensor to the train body, calibrate the binocular camera and lidar device, and calibrate the internal and external parameters of the binocular camera and the external parameters of the lidar.

[0068] The external parameters of the binocular camera and lidar equipment include the transformation matrix between the lidar coordinate system and the train body coordinate system, and the transformation matrix between the binocular camera coordinate system and the lidar coordinate system; the internal parameters of the binocular camera are the transformation matrix between pixel coordinates and the unit sphere of the binocular camera coordinate system.

[0069] S2. When the train is running, it continuously collects image data along the railway line and detects and identifies the virtual transponder numbers printed on the trackside signs.

[0070] Specifically, the railway line image data continuously collected by the binocular camera device is used to detect signs, obtain target areas, and extract identification information, which is the virtual transponder number printed on the trackside sign.

[0071] The image data along the railway line includes two image frames captured by the two cameras of the binocular camera device.

[0072] S3. Align the image point cloud with the data collected by the binocular camera and lidar equipment, and calculate the distance between the train and the trackside sign and the first confidence interval based on the point cloud information of the target area.

[0073] Image point cloud alignment comprises two parts: spatial and temporal alignment of the image point cloud data. Spatial alignment includes the following steps: projecting the pixel coordinates of the target region onto a unit sphere of the binocular camera coordinate system using the intrinsic parameters of the binocular camera, based on the camera model; transforming the target region from the camera coordinate system to the LiDAR coordinate system using the extrinsic parameters of the binocular camera to obtain the target point cloud region; and temporal alignment, considering the sampling time difference between the image frame and the point cloud frame, correcting the target point cloud region.

[0074] S4. Based on the extracted identification information, the distance between the train and the trackside sign and the first confidence interval are converted into the train's one-dimensional on-track position and the second confidence interval. The train's passing time is estimated, and a virtual transponder is triggered when the train passes the trackside sign. Otherwise, return to S2 to re-detect the sign.

[0075] Among them, the outer points of the above-mentioned distance measurement between the train and the trackside sign were eliminated based on geometric constraints;

[0076] The virtual transponder is triggered by sending a communication frame to the on-board equipment main control unit. The communication frame includes the timestamp of the train passing the trackside sign and the maximum and minimum safe front-end positions of the train at the time of passing.

[0077] Specifically, in S4, the train, based on the trackside sign mentioned above, determines the time from the initial recognition of the trackside sign to its passing, going through multiple S2 to S4 cycles. Based on the wheel axle speed sensor input and the aforementioned distance measurement between the train and the trackside sign, a state equation is constructed using the on-track distance between the train and the virtual transponder as the state variable, and then estimated and corrected. Until S2 no longer detects the trackside sign, the time when the train passed the trackside sign is estimated based on the state variable, and the result is reported to the virtual transponder.

[0078] Specifically, detecting and identifying virtual transponder numbers on trackside signs includes:

[0079] The target detection network performs sign detection on continuous image frames along the railway line captured by the two cameras of the binocular camera. When the confidence scores at the outputs of both detection networks exceed a threshold, text recognition is performed on the target regions output by both detection networks. If the text information recognized in the two target regions is consistent and the confidence scores both exceed the threshold, the virtual transponder number is obtained.

[0080] Specifically, it includes two parts: a trackside sign detection network based on convolutional neural networks and an OCR character recognition network based on recurrent neural networks or Transformers.

[0081] Specifically, a trackside sign detection network based on convolutional neural networks is constructed using a supervised learning-based target detection method to detect target regions containing trackside signs;

[0082] The method employs a convolutional neural network (CNN) to detect image frames and identify target regions containing signs. These target regions are defined as image areas whose detection confidence exceeds a threshold, as outlined by the output of the CNN. The supervised learning-based target detection method requires constructing a dataset containing visual images of the target and corresponding annotation information, pre-training the CNN parameters, and obtaining a target detection network model. The target detection network includes two types: anchor-bound and non-anchor-bound. Anchor-bound networks further include single-stage and two-stage detection methods.

[0083] Specifically, the trackside sign detection network adopts the YOLOv5 network model, consisting of four parts: input, backbone, neck, and output. The specific target detection process is as follows: Figure 4 As shown: The input end scales the image frames input from the binocular camera to a specified size and performs normalization processing; the backbone fully extracts image features through the network skeleton to obtain feature maps; the neck combines feature pyramids and path aggregation to fuse feature maps of different scales and detect targets of different sizes; the output end uses the GLOU_Loss loss function to perform bounding box regression, resulting in:

[0084]

[0085] Among them, A c Let U be the area of ​​the minimum bounding box containing the two target boxes, and U be the intersection of the two target boxes. c –U represents the area of ​​the minimum bounding box after removing the two target boxes.

[0086] Specifically, OCR character recognition technology is used to identify the target area to obtain the virtual transponder number;

[0087] The identification information extraction adopts OCR character recognition technology, including a hybrid model that combines a visual model with a sequence model for sequential modeling, and an OCR character recognition network that combines a single visual model with a Transformer.

[0088] Specifically, the text recognition uses the Transformer-based SVTR algorithm. The target region output by the target detection network passes through a three-stage height-decreasing network, with each stage performing merging and combination operations. Finally, it is recognized through linear prediction to obtain the virtual transponder number.

[0089] Specifically, from spatial and temporal dimensions, image point cloud alignment is performed on the railway line image data and the railway line point cloud data, including:

[0090] From a spatial perspective, the pixel coordinates of the target area are projected onto the unit sphere of the binocular camera coordinate system based on the intrinsic parameters of the binocular camera; the target area is then transformed from the binocular camera coordinate system to the lidar coordinate system based on the extrinsic parameters of the binocular camera to obtain the target point cloud region.

[0091] From a temporal perspective, the image data and point cloud data along the railway line are time-aligned. Based on the sampling time difference between the image frame and the point cloud frame, the target point cloud region is corrected to obtain the point cloud information of the corrected target region.

[0092] Specifically, image data along the railway line is acquired through the binocular camera.

[0093] Specifically, based on the measurement principle of the lidar device, before the image point cloud is aligned, the point cloud data is projected to the end of the scan to eliminate point cloud motion distortion.

[0094] Specifically, the calculation of the distance from the train to the trackside sign and the first confidence interval includes:

[0095] The point cloud information of the target area is filtered by reflection intensity, and the coordinates of the filtered points are formed into a matrix.

[0096]

[0097] Based on the plane equation, a least squares problem is constructed. This least squares problem is solved using QR decomposition. Plane fitting is then performed on the selected point set to obtain:

[0098]

[0099] Where, x i y i z i For point p i Coordinates, v is the normal vector of the fitted plane, and d is the plane fitting error;

[0100] Calculate the centroid coordinates of a planar point cloud

[0101]

[0102] The centroid coordinates of the planar point cloud represent the position of the trackside sign, and the distance between the train and the trackside sign is obtained.

[0103] The first confidence interval is determined based on the inherent ranging error, plane fitting error, centroid calculation error of the lidar equipment, and the installation error of the trackside sign in the direction perpendicular to the track line.

[0104] Specifically, point cloud data along the railway line is acquired through the lidar device.

[0105] This invention, in its embodiments, aligns the image point cloud data collected by the binocular camera and lidar device from both spatial and temporal dimensions to obtain a corrected target point cloud region. The point cloud within this region is then filtered by reflection intensity, and a plane fit is performed on the filtered point cloud. The centroid coordinates of the planar point cloud are calculated and used to represent the position of the trackside sign. The distance between the train and the trackside sign, as well as the first confidence interval, are then calculated.

[0106] In one specific embodiment of the present invention, the spatial dimension of the image point cloud alignment involves projecting the pixel coordinates of the corner points of the target region onto the unit sphere of the binocular camera coordinate system based on the camera model and the camera intrinsic parameters calibrated in S1.

[0107]

[0108] Where K represents the intrinsic parameters of the calibrated stereo camera; u and v are the pixel coordinates of the corner points of the target region bounding box; X, Y, Z, and P represent the coordinates of the corner points of the target region bounding box in the camera coordinate system; f x f y c represents the camera's focal length along the x and y axes, respectively; x c y The coordinates of the camera's optical center;

[0109] Next, the corner points of the target area are transformed from the camera coordinate system to the lidar coordinate system. The transformation formula is as follows:

[0110]

[0111] Where [R|t] represents the extrinsic parameters of the calibrated binocular camera in S1, R lc P is the rotation matrix between the camera coordinate system and the lidar coordinate system, achieving spatial alignment. t represents the displacement between the two coordinate systems. c P l These represent the coordinates of the corner points of the target area frame in the camera coordinate system and the lidar coordinate system, respectively.

[0112] Specifically, image point cloud time alignment takes into account the sampling time difference between image frames and point cloud frames, corrects the target area output by the image frame, and expands it to the union of the target areas at the two sampling times.

[0113] In one specific embodiment of the present invention, based on the sampling time difference between the image frame and the point cloud frame, the spatially aligned target region is converted from the image frame sampling time to the point cloud frame sampling time, and then expanded to the union of the target regions at the two sampling times, thus completing the alignment in the temporal dimension. Taking the corner point of the target region as an example, Figure 5 and Figure 6 These are schematic diagrams illustrating the changes in the horizontal and vertical angles at two sampling times, where T... c T l Let L be the train position at the sampling time of the image frame and the point cloud frame, α and β be the horizontal and vertical angles at the sampling time of the image frame, respectively, and Δα and Δβ be the changes in the corresponding directional angles at the sampling time of the point cloud frame. c L l L ch L lh The distance measurement at the corresponding moment and the horizontal component of the distance, where a and d are the width of the sign and the horizontal distance from the track, h is the vertical distance between the sampling point and the horizontal plane of the lidar, and L o Let be the displacement of the train between two time points. The displacement can be obtained from the model measured by the wheel-axle speed sensor:

[0114]

[0115] Where n is the number of pulses at the corresponding moment, D is the wheel diameter of the train, and N is the number of teeth on the wheel axle speed sensor. For example... Figure 5 As shown, according to the Law of Cosines, we can obtain:

[0116]

[0117] According to geometric relationships, we know that:

[0118]

[0119] Substituting the values, we obtain the change in the horizontal angle at the sampling time of the point cloud frame as follows:

[0120]

[0121] like Figure 6 As shown, according to the Law of Cosines, we can obtain:

[0122]

[0123] Where c represents the sampling time of the point cloud frame, and the distance between the intersection point of the ray formed by the sampling point and the perpendicular line of the lidar plane and the lidar perpendicularly at an angle of β; e represents the distance between the above intersection point and the sampling point.

[0124] Based on geometric relationships, c, e, and L1 can be obtained as follows:

[0125]

[0126] e = (L ch -L lh )tanβ;

[0127]

[0128] Substituting the values, we obtain the change in the vertical angle at the sampling time of the point cloud frame as follows:

[0129]

[0130] Finally, the target region at the sampling time of the corrected point cloud frame is:

[0131] [min(α i ,α i +Δα i ),max(α i ,α i +Δa i )];

[0132] [min(β,β i +Δβ i ), max(β) i ,β i +Δβ i )];

[0133] Where α and β are the angles between the horizontal and vertical directions of the lidar polar coordinates, respectively, and i is the corner point of the target area at the sampling time.

[0134] Specifically, in step three, the distance calculation between the train and the trackside sign is as follows: Figure 7 As shown, the process includes the following steps: filtering the reflection intensity of the target point cloud region; performing planar fitting on the filtered point cloud based on the actual size of the trackside sign; calculating the centroid coordinates of the planar point cloud and using them to represent the position of the trackside sign, obtaining the distance between the train and the trackside sign, and calculating the first confidence interval of the distance.

[0135] Specifically, the first confidence interval consists of the following four parts: the inherent ranging error of the lidar equipment, the plane fitting error, the centroid calculation error, and the installation error of the trackside sign in the direction perpendicular to the track line.

[0136] In this embodiment of the invention, the distance between the train and the trackside sign and the first confidence interval are converted into the distance between the train and the virtual transponder represented by the trackside sign and the second confidence interval. The track straightness assumption error and the installation deviation of the trackside sign along the track direction are added to the second confidence interval.

[0137] Specifically, based on the virtual transponder number, the on-track position of the virtual transponder is obtained from the electronic map, and combined with the distance between the train and the virtual transponder, the on-track position of the train is obtained.

[0138] Specifically, based on the train operation pulse data and the distance between the train and the trackside sign;

[0139] A state equation is constructed using the on-track distance between the train and the virtual transponder as the state variable to predict the on-track distance between the train and the virtual transponder; at the same time, the first confidence interval is converted into a second confidence interval.

[0140] And trigger the virtual transponder when the state vector is less than zero and the trackside sign detection fails;

[0141] Wherein, the first confidence interval is the confidence interval for measuring the distance between the train and the trackside sign;

[0142] The second confidence interval is the confidence interval of the on-orbit distance between the train and the virtual transponder.

[0143] In this embodiment of the invention, a state equation is constructed to predict the on-track distance between the train and the virtual transponder, and the distance measurement between the train and the trackside sign in step S3 is fused to obtain the optimal estimate of the on-track distance between the train and the virtual transponder. Simultaneously, the first confidence interval of the distance measurement between the train and the trackside sign in step S3 is converted into a second confidence interval for the on-track distance between the train and the virtual transponder, and the virtual transponder is triggered when the state vector is less than zero and the sign detection in step two fails. Otherwise, the process returns to step S2.

[0144] Specifically, under the assumption of a straight trajectory, the state equation is constructed as follows:

[0145]

[0146] Where, x k Let u be the on-orbit distance between the train and the virtual transponder at time k. k The train travel distance calculated by the wheel speed sensor, y k Let α be the distance between the train and the trackside sign, and β be the angles between the horizontal and vertical directions of the laser radar device, respectively. k n k These are the input noise of the wheel axle speed sensor and the measurement noise of the train's one-dimensional on-track position, respectively.

[0147] Specifically, the optimal estimate of the state equation is obtained by Kalman filtering, wherein the prediction step is calculated as follows:

[0148]

[0149] Where Q represents the noise of the wheel axle speed sensor;

[0150] The Kalman gain is calculated as follows:

[0151]

[0152] Where R is the noise level measured between the train and the trackside sign.

[0153] The updated state vector and confidence are obtained as follows:

[0154]

[0155] In the formula, This is the prior estimate of the state vector at time k; The prior covariance matrix is ​​divided into the state vectors at time k. The posterior estimates of the state vectors at times k-1 and k are respectively. The posterior covariance matrices of the state vectors at times k-1 and k, respectively; K k Let K be the Kalman gain at time k.

[0156] Specifically, the second confidence interval consists of two parts: a random part and a fixed part;

[0157] The random part is the confidence level obtained by solving the state equation to get the state vector;

[0158] The fixed components include the assumed error of the track straightness and the installation deviation of the trackside signs along the track direction.

[0159] Figure 8 This is a schematic diagram of the error portion of the track straightness assumption. The on-track position is calculated using the minimum radius of curvature of the track as the maximum value of this portion, and the on-track position is calculated using the negative value of the minimum radius of curvature as the minimum value of this portion. Based on... Figure 8 Based on the geometric relationships shown, the error portion of the trajectory straight-line assumption can be derived as follows:

[0160]

[0161] In the formula, r is the minimum radius of curvature of the track, α and β are the angles between the horizontal and vertical directions of the laser radar device in polar coordinates, and L represents the distance between the train and the trackside sign.

[0162] Finally, the second confidence interval is expressed as:

[0163]

[0164] In the formula, dev is the installation deviation of the trackside sign along the track direction, and P is the confidence level of the state vector obtained by solving the state equation.

[0165] Specifically, the virtual transponder is triggered by establishing a communication connection with the on-board equipment main control unit via an RS422 serial port, and sending communication frames at a period of hundreds of milliseconds. The communication frame includes a timestamp of the train passing the trackside sign, the maximum and minimum safe leading-edge positions of the train at the time of passing, the confidence level of the trackside sign, and the identified virtual transponder number, and the correctness of the bidirectional data is ensured by CRC check.

[0166] In one specific embodiment of the present invention, a train positioning system, such as... Figure 1 As shown, it includes:

[0167] The sensor module is used to collect image data, point cloud data, and train operation pulse data along the railway line.

[0168] Electronic maps are used to store track geographic information and virtual transponder data required for train positioning;

[0169] The integrated positioning unit is used to receive and process the railway line image data, railway line point cloud data and train operation pulse data, obtain the distance between the train and the trackside sign, and trigger the virtual transponder when passing the trackside sign.

[0170] Among them, the trackside signs are set up next to the train tracks as indicators for the line's virtual transponders.

[0171] The sensor module includes a binocular camera device, a lidar device, and a wheel speed sensor.

[0172] A binocular camera device is used to collect image data along the railway line and send it to the integrated positioning unit when the train is running; a lidar device is used to acquire point cloud data along the railway line and send it to the integrated positioning unit when the train is running; wheel speed sensors are installed at different axle ends of the train and send the pulse data of the train's operation to the integrated positioning unit when the train is running.

[0173] like Figure 2 The diagram shown is an installation schematic of the binocular camera and lidar device in an embodiment of the present invention. The binocular camera device is installed on the central axis of the top of the train head to ensure that the viewing angles of the cameras on both sides of the train are the same. It is used to collect image data along the railway line and send it to the integrated positioning unit when the train is running. The lidar device is installed on the central axis of the top of the train head to ensure that the lidar scanning ranges on both sides of the train are the same. It is used to acquire point cloud information along the railway line and send it to the integrated positioning unit when the train is running.

[0174] The electronic map is used to store the one-to-one mapping relationship between the virtual transponders represented by the trackside signs and their one-dimensional on-track positions. After the virtual transponder is triggered, the on-board equipment main control unit obtains the train's on-track position and transponder messages by querying the electronic map.

[0175] The integrated positioning unit is connected to the wheel speed sensor, binocular camera, lidar equipment, and the main control unit of the existing train control on-board equipment. It is used to receive and comprehensively process images, point cloud information and train travel distance information along the railway line, obtain the distance between the train and the trackside sign, and report the virtual transponder to the main control unit of the on-board equipment when passing the trackside sign.

[0176] Specifically, the trackside sign is made of 1.2mm thick aluminum plate, measuring 50cm × 40cm. The numbers are printed with Arial font, and a high-quality reflective film is applied to the surface to ensure good reflection of auxiliary light sources in low-light conditions. The sign is printed with virtual transponder numbers and is positioned at fixed intervals perpendicular to the track line.

[0177] Specifically, the electronic map includes track geographic information and transponder data required for train positioning. It uses the station, track, and sequence position to form a unique identifier for the transponder and stores its on-track position and transponder messages.

[0178] Specifically, the integrated positioning unit includes a sign visual detection module, an image point cloud joint ranging module, and an information conversion module. The sign visual detection module performs visual target detection on the trackside sign, performs text recognition on the target area, and extracts the virtual transponder number. The image point cloud joint ranging module performs image point cloud registration on the output information of the binocular camera and lidar equipment, detects and compensates for errors in the point cloud information of the trackside sign area, and finally calculates the distance between the train and the trackside sign and a first confidence interval. The information conversion module converts the distance between the train and the trackside sign and the first confidence interval into the train's one-dimensional coordinate system position on track and a second confidence interval. When it determines that the train has passed the trackside sign, it reports the virtual transponder to the onboard main control unit.

[0179] The train positioning system of this invention can be applied to CTCS2 level, CTCS3 level and new train control systems, and achieves absolute on-track positioning of the train by triggering a virtual transponder.

[0180] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0181] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A train positioning method, characterized in that, include: Continuously collect train operation data, which includes railway line image data, railway line point cloud data, and train operation pulse data; Based on the image data along the railway line, the virtual transponder numbers on the trackside signs are detected and identified. From the spatial and temporal dimensions, the image data and point cloud data along the railway line are aligned to obtain the point cloud information of the corrected target area. Based on the point cloud information of the corrected target area, the distance from the train to the trackside sign and the first confidence interval are calculated. Based on electronic map data, the distance from the train to the trackside sign and the first confidence interval are converted into the train's on-track position and the second confidence interval, the train's passing time is predicted, and a virtual transponder is triggered when the train passes the trackside sign; otherwise, the trackside sign detection is repeated. Based on the virtual transponder number, the on-track position of the virtual transponder is obtained from the electronic map, and combined with the distance between the train and the virtual transponder, the on-track position of the train is obtained. Based on the train operation pulse data and the distance between the train and the trackside sign; A state equation is constructed using the on-track distance between the train and the virtual transponder as the state variable to predict the on-track distance between the train and the virtual transponder; at the same time, the first confidence interval is converted into a second confidence interval. And trigger the virtual transponder when the state vector is less than zero and the trackside sign detection fails; Wherein, the first confidence interval is the confidence interval for measuring the distance between the train and the trackside sign; The second confidence interval is the confidence interval of the on-orbit distance between the train and the virtual transponder; The second confidence interval consists of two parts: a random part and a fixed part. The random part is the confidence level obtained by solving the state equation to get the state vector; The fixed components include the track straightness assumption error and the installation deviation of the trackside sign along the track direction; The error portion of the trajectory straight line assumption is as follows: ； In the formula, r is the minimum radius of curvature of the track, α and β are the angles between the horizontal and vertical directions of the laser radar device in polar coordinates, and L represents the distance between the train and the trackside sign. Finally, the second confidence interval is expressed as: ； In the formula, dev tan Let P be the installation deviation of the trackside sign along the track direction, and let P be the confidence level of the state vector obtained by solving the state equation. Let be the posterior covariance matrix of the state vector at time k.

2. The train positioning method according to claim 1, characterized in that, Detecting and identifying virtual transponder numbers on trackside signs, including: A supervised learning-based target detection method is used to detect target areas containing trackside signs. The target area is identified using OCR character recognition technology to obtain the virtual transponder number.

3. The train positioning method according to claim 1, characterized in that, Image point cloud alignment is performed on the railway line image data and the railway line point cloud data from both spatial and temporal dimensions, including: From a spatial perspective, the pixel coordinates of the target area are projected onto the unit sphere of the binocular camera coordinate system based on the intrinsic parameters of the binocular camera; the target area is then transformed from the binocular camera coordinate system to the lidar coordinate system based on the extrinsic parameters of the binocular camera to obtain the target point cloud region. From a temporal perspective, the image data and point cloud data along the railway line are time-aligned. Based on the sampling time difference between the image frame and the point cloud frame, the target point cloud region is corrected to obtain the point cloud information of the corrected target region. Specifically, image data along the railway line is acquired through the binocular camera.

4. The train positioning method according to claim 1, characterized in that, Calculate the distance from the train to the trackside sign and the first confidence interval, including: The point cloud information of the target area is filtered by reflection intensity, and the filtered point cloud information is fitted by plane in combination with the actual size of the trackside sign. Calculate the centroid coordinates of the planar point cloud, where the centroid coordinates of the planar point cloud represent the position of the trackside sign, and obtain the distance between the train and the trackside sign; The first confidence interval is determined based on the inherent ranging error, plane fitting error, centroid calculation error of the lidar equipment, and the installation error of the trackside sign in the direction perpendicular to the track line. Specifically, point cloud data along the railway line is acquired through the lidar device.

5. The train positioning method according to claim 1, characterized in that, The state equation is expressed as: ； Where, x k Let u be the on-orbit distance between the train and the virtual transponder at time k. k y represents the distance the train travels. k Let α be the distance between the train and the trackside sign, and β be the angles between the horizontal and vertical directions of the laser radar device, respectively. k n k These are the input noise from the wheel axle speed sensor and the measurement noise from the train's on-track position, respectively.

6. The train positioning method according to claim 5, characterized in that, The optimal estimate of the state equation is obtained by Kalman filtering, wherein the prediction step is calculated as follows: ； in, Q k The noise of the wheel axle speed sensor at time k; The Kalman gain is calculated as follows: ； in, R k The noise level is the measurement noise of the distance between the train and the trackside sign at time k. The updated state vector and confidence scores are obtained as follows: ； In the formula, Let be the prior covariance matrix of the state vector at time k; Let be the posterior covariance matrix of the state vector at time k-1; This is the prior estimate of the state vector at time k; This is the posterior estimate of the state vector at time k-1; This is the posterior estimate of the state vector at time k; The Kalman gain at time k; Let be the posterior covariance matrix of the state vector at time k.

7. A train positioning system, employing the train positioning method according to any one of claims 1-6, characterized in that, include: The sensor module is used to collect image data, point cloud data, and train operation pulse data along the railway line. Electronic maps are used to store track geographic information and virtual transponder data required for train positioning; The integrated positioning unit is used to receive and process the railway line image data, railway line point cloud data and train operation pulse data, obtain the distance between the train and the trackside sign, and trigger the virtual transponder when passing the trackside sign.

Images